Julian Martin Eisenschlos

Julian Martin Eisenschlos, Jeremy R. Cole, Fangyu Liu et al. · deepmind

We introduce a new in-context learning paradigm to measure Large Language Models' (LLMs) ability to learn novel words during inference. In particular, we rewrite Winograd-style co-reference resolution problems by replacing the key concept word with a synthetic but plausible word that the model must understand to complete the task. Solving this task requires the model to make use of the dictionary definition of the new word given in the prompt. This benchmark addresses word acquisition, one important aspect of the diachronic degradation known to afflict LLMs. As LLMs are frozen in time at the moment they are trained, they are normally unable to reflect the way language changes over time. We show that the accuracy of LLMs compared to the original Winograd tasks decreases radically in our benchmark, thus identifying a limitation of current models and providing a benchmark to measure future improvements in LLMs ability to do in-context learning.

Fangyu Liu, Julian Martin Eisenschlos, Francesco Piccinno et al. · deepmind

Visual language such as charts and plots is ubiquitous in the human world. Comprehending plots and charts requires strong reasoning skills. Prior state-of-the-art (SOTA) models require at least tens of thousands of training examples and their reasoning capabilities are still much limited, especially on complex human-written queries. This paper presents the first one-shot solution to visual language reasoning. We decompose the challenge of visual language reasoning into two steps: (1) plot-to-text translation, and (2) reasoning over the translated text. The key in this method is a modality conversion module, named as DePlot, which translates the image of a plot or chart to a linearized table. The output of DePlot can then be directly used to prompt a pretrained large language model (LLM), exploiting the few-shot reasoning capabilities of LLMs. To obtain DePlot, we standardize the plot-to-table task by establishing unified task formats and metrics, and train DePlot end-to-end on this task. DePlot can then be used off-the-shelf together with LLMs in a plug-and-play fashion. Compared with a SOTA model finetuned on more than >28k data points, DePlot+LLM with just one-shot prompting achieves a 24.0% improvement over finetuned SOTA on human-written queries from the task of chart QA.

Fangyu Liu, Francesco Piccinno, Syrine Krichene et al. · deepmind

Visual language data such as plots, charts, and infographics are ubiquitous in the human world. However, state-of-the-art vision-language models do not perform well on these data. We propose MatCha (Math reasoning and Chart derendering pretraining) to enhance visual language models' capabilities in jointly modeling charts/plots and language data. Specifically, we propose several pretraining tasks that cover plot deconstruction and numerical reasoning which are the key capabilities in visual language modeling. We perform the MatCha pretraining starting from Pix2Struct, a recently proposed image-to-text visual language model. On standard benchmarks such as PlotQA and ChartQA, the MatCha model outperforms state-of-the-art methods by as much as nearly 20%. We also examine how well MatCha pretraining transfers to domains such as screenshots, textbook diagrams, and document figures and observe overall improvement, verifying the usefulness of MatCha pretraining on broader visual language tasks.

Fangyu Liu, Julian Martin Eisenschlos, Jeremy R. Cole et al. · deepmind

Language models (LMs) trained on raw texts have no direct access to the physical world. Gordon and Van Durme (2013) point out that LMs can thus suffer from reporting bias: texts rarely report on common facts, instead focusing on the unusual aspects of a situation. If LMs are only trained on text corpora and naively memorise local co-occurrence statistics, they thus naturally would learn a biased view of the physical world. While prior studies have repeatedly verified that LMs of smaller scales (e.g., RoBERTa, GPT-2) amplify reporting bias, it remains unknown whether such trends continue when models are scaled up. We investigate reporting bias from the perspective of colour in larger language models (LLMs) such as PaLM and GPT-3. Specifically, we query LLMs for the typical colour of objects, which is one simple type of perceptually grounded physical common sense. Surprisingly, we find that LLMs significantly outperform smaller LMs in determining an object's typical colour and more closely track human judgments, instead of overfitting to surface patterns stored in texts. This suggests that very large models of language alone are able to overcome certain types of reporting bias that are characterized by local co-occurrences.

Ewa Andrejczuk, Julian Martin Eisenschlos, Francesco Piccinno et al.

Encoder-only transformer models have been successfully applied to different table understanding tasks, as in TAPAS (Herzig et al., 2020). A major limitation of these architectures is that they are constrained to classification-like tasks such as cell selection or entailment detection. We present TABT5, an encoder-decoder model that generates natural language text based on tables and textual inputs. TABT5 overcomes the encoder-only limitation by incorporating a decoder component and leverages the input structure with table specific embeddings and pre-training. TABT5 achieves new state-of-the-art results on several domains, including spreadsheet formula prediction with a 15% increase in sequence accuracy, QA with a 2.5% increase in sequence accuracy and data-to-text generation with a 2.5% increase in BLEU.

Iulia-Maria Comsa, Julian Martin Eisenschlos, Srini Narayanan

We propose a benchmark to assess the capability of large language models to reason with conventional metaphors. Our benchmark combines the previously isolated topics of metaphor detection and commonsense reasoning into a single task that requires a model to make inferences by accurately selecting between the literal and metaphorical register. We examine the performance of state-of-the-art pre-trained models on binary-choice tasks and find a large discrepancy between the performance of small and very large models, going from chance to near-human level. We also analyse the largest model in a generative setting and find that although human performance is approached, careful multiple-shot prompting is required.

Jeremy R. Cole, Palak Jain, Julian Martin Eisenschlos et al.

Identifying the difference between two versions of the same article is useful to update knowledge bases and to understand how articles evolve. Paired texts occur naturally in diverse situations: reporters write similar news stories and maintainers of authoritative websites must keep their information up to date. We propose representing factual changes between paired documents as question-answer pairs, where the answer to the same question differs between two versions. We find that question-answer pairs can flexibly and concisely capture the updated contents. Provided with paired documents, annotators identify questions that are answered by one passage but answered differently or cannot be answered by the other. We release DIFFQG which consists of 759 QA pairs and 1153 examples of paired passages with no factual change. These questions are intended to be both unambiguous and information-seeking and involve complex edits, pushing beyond the capabilities of current question generation and factual change detection systems. Our dataset summarizes the changes between two versions of the document as questions and answers, studying automatic update summarization in a novel way.

Aashna Jena, Vivek Gupta, Manish Shrivastava et al.

Creating challenging tabular inference data is essential for learning complex reasoning. Prior work has mostly relied on two data generation strategies. The first is human annotation, which yields linguistically diverse data but is difficult to scale. The second category for creation is synthetic generation, which is scalable and cost effective but lacks inventiveness. In this research, we present a framework for semi-automatically recasting existing tabular data to make use of the benefits of both approaches. We utilize our framework to build tabular NLI instances from five datasets that were initially intended for tasks like table2text creation, tabular Q/A, and semantic parsing. We demonstrate that recasted data could be used as evaluation benchmarks as well as augmentation data to enhance performance on tabular NLI tasks. Furthermore, we investigate the effectiveness of models trained on recasted data in the zero-shot scenario, and analyse trends in performance across different recasted datasets types.

Negar Foroutan, Angelika Romanou, Matin Ansaripour et al.

Documents are fundamental to preserving and disseminating information, often incorporating complex layouts, tables, and charts that pose significant challenges for automatic document understanding (DU). While vision-language large models (VLLMs) have demonstrated improvements across various tasks, their effectiveness in processing long-context vision inputs remains unclear. This paper introduces WikiMixQA, a benchmark comprising 1,000 multiple-choice questions (MCQs) designed to evaluate cross-modal reasoning over tables and charts extracted from 4,000 Wikipedia pages spanning seven distinct topics. Unlike existing benchmarks, WikiMixQA emphasizes complex reasoning by requiring models to synthesize information from multiple modalities. We evaluate 12 state-of-the-art vision-language models, revealing that while proprietary models achieve ~70% accuracy when provided with direct context, their performance deteriorates significantly when retrieval from long documents is required. Among these, GPT-4-o is the only model exceeding 50% accuracy in this setting, whereas open-source models perform considerably worse, with a maximum accuracy of 27%. These findings underscore the challenges of long-context, multi-modal reasoning and establish WikiMixQA as a crucial benchmark for advancing document understanding research.

Zilong Wang, Hao Zhang, Chun-Liang Li et al.

Table-based reasoning with large language models (LLMs) is a promising direction to tackle many table understanding tasks, such as table-based question answering and fact verification. Compared with generic reasoning, table-based reasoning requires the extraction of underlying semantics from both free-form questions and semi-structured tabular data. Chain-of-Thought and its similar approaches incorporate the reasoning chain in the form of textual context, but it is still an open question how to effectively leverage tabular data in the reasoning chain. We propose the Chain-of-Table framework, where tabular data is explicitly used in the reasoning chain as a proxy for intermediate thoughts. Specifically, we guide LLMs using in-context learning to iteratively generate operations and update the table to represent a tabular reasoning chain. LLMs can therefore dynamically plan the next operation based on the results of the previous ones. This continuous evolution of the table forms a chain, showing the reasoning process for a given tabular problem. The chain carries structured information of the intermediate results, enabling more accurate and reliable predictions. Chain-of-Table achieves new state-of-the-art performance on WikiTQ, FeTaQA, and TabFact benchmarks across multiple LLM choices.

Julian Martin Eisenschlos, Bhuwan Dhingra, Jannis Bulian et al.

We release FoolMeTwice (FM2 for short), a large dataset of challenging entailment pairs collected through a fun multi-player game. Gamification encourages adversarial examples, drastically lowering the number of examples that can be solved using "shortcuts" compared to other popular entailment datasets. Players are presented with two tasks. The first task asks the player to write a plausible claim based on the evidence from a Wikipedia page. The second one shows two plausible claims written by other players, one of which is false, and the goal is to identify it before the time runs out. Players "pay" to see clues retrieved from the evidence pool: the more evidence the player needs, the harder the claim. Game-play between motivated players leads to diverse strategies for crafting claims, such as temporal inference and diverting to unrelated evidence, and results in higher quality data for the entailment and evidence retrieval tasks. We open source the dataset and the game code.

Sebastian Prillo, Julian Martin Eisenschlos

While sorting is an important procedure in computer science, the argsort operator - which takes as input a vector and returns its sorting permutation - has a discrete image and thus zero gradients almost everywhere. This prohibits end-to-end, gradient-based learning of models that rely on the argsort operator. A natural way to overcome this problem is to replace the argsort operator with a continuous relaxation. Recent work has shown a number of ways to do this, but the relaxations proposed so far are computationally complex. In this work we propose a simple continuous relaxation for the argsort operator which has the following qualities: it can be implemented in three lines of code, achieves state-of-the-art performance, is easy to reason about mathematically - substantially simplifying proofs - and is faster than competing approaches. We open source the code to reproduce all of the experiments and results.

Mubashara Akhtar, Chenxi Pang, Andreea Marzoca et al. · deepmind

Language models, potentially augmented with tool usage such as retrieval are becoming the go-to means of answering questions. Understanding and answering questions in real-world settings often requires retrieving information from different sources, processing and aggregating data to extract insights, and presenting complex findings in form of structured artifacts such as novel tables, charts, or infographics. In this paper, we introduce TANQ, the first open domain question answering dataset where the answers require building tables from information across multiple sources. We release the full source attribution for every cell in the resulting table and benchmark state-of-the-art language models in open, oracle, and closed book setups. Our best-performing baseline, Gemini Flash reaches an overall F1 score of 60.7, lagging behind human performance by 12.3 points. We analyse baselines' performance across different dataset attributes such as different skills required for this task, including multi-hop reasoning, math operations, and unit conversions. We further discuss common failures in model-generated answers, suggesting that TANQ is a complex task with many challenges ahead.

Hernán Maina, Guido Ivetta, Mateo Lione Stuto et al.

Visually impaired people could benefit from Visual Question Answering (VQA) systems to interpret text in their surroundings. However, current models often struggle with recognizing text in the photos taken by this population. Through in-depth interviews with visually impaired individuals, we identified common framing conventions that frequently result in misaligned text. Existing VQA benchmarks primarily feature well-oriented text captured by sighted users, under-representing these challenges. To address this gap, we introduce ROtated SAmpling (ROSA), a decoding strategy that enhances VQA performance in text-rich images with incorrectly oriented text. ROSA outperforms Greedy decoding by 11.7 absolute points in the best-performing model.

Julian Martin Eisenschlos, Hernán Maina, Guido Ivetta et al.

Recently, large multi-modal models (LMMs) have emerged with the capacity to perform vision tasks such as captioning and visual question answering (VQA) with unprecedented accuracy. Applications such as helping the blind or visually impaired have a critical need for precise answers. It is specially important for models to be well calibrated and be able to quantify their uncertainty in order to selectively decide when to answer and when to abstain or ask for clarifications. We perform the first in-depth analysis of calibration methods and metrics for VQA with in-context learning LMMs. Studying VQA on two answerability benchmarks, we show that the likelihood score of visually grounded models is better calibrated than in their text-only counterparts for in-context learning, where sampling based methods are generally superior, but no clear winner arises. We propose Avg BLEU, a calibration score combining the benefits of both sampling and likelihood methods across modalities.

Xingchen Wan, Ruoxi Sun, Hootan Nakhost et al.

A hallmark of modern large language models (LLMs) is their impressive general zero-shot and few-shot abilities, often elicited through in-context learning (ICL) via prompting. However, while highly coveted and being the most general, zero-shot performances in LLMs are still typically weaker due to the lack of guidance and the difficulty of applying existing automatic prompt design methods in general tasks when ground-truth labels are unavailable. In this study, we address this by presenting Universal Self-Adaptive Prompting (USP), an automatic prompt design approach specifically tailored for zero-shot learning (while compatible with few-shot). Requiring only a small amount of unlabeled data and an inference-only LLM, USP is highly versatile: to achieve universal prompting, USP categorizes a possible NLP task into one of the three possible task types and then uses a corresponding selector to select the most suitable queries and zero-shot model-generated responses as pseudo-demonstrations, thereby generalizing ICL to the zero-shot setup in a fully automated way. We evaluate USP with PaLM and PaLM 2 models and demonstrate performances that are considerably stronger than standard zero-shot baselines and often comparable to or even superior to few-shot baselines across more than 40 natural language understanding, natural language generation, and reasoning tasks.

Jeremy R. Cole, Michael J. Q. Zhang, Daniel Gillick et al.

Trustworthy language models should abstain from answering questions when they do not know the answer. However, the answer to a question can be unknown for a variety of reasons. Prior research has focused on the case in which the question is clear and the answer is unambiguous but possibly unknown, but the answer to a question can also be unclear due to uncertainty of the questioner's intent or context. We investigate question answering from this perspective, focusing on answering a subset of questions with a high degree of accuracy, from a set of questions in which many are inherently ambiguous. In this setting, we find that the most reliable approach to decide when to abstain involves quantifying repetition within sampled model outputs, rather than the model's likelihood or self-verification as used in prior work. We find this to be the case across different types of uncertainty and model scales,and with or without instruction tuning. Our results suggest that sampling-based confidence scores help calibrate answers to relatively unambiguous questions, with more dramatic improvements on ambiguous questions.

Julian Martin Eisenschlos, Maharshi Gor, Thomas Müller et al.

This work presents a sparse-attention Transformer architecture for modeling documents that contain large tables. Tables are ubiquitous on the web, and are rich in information. However, more than 20% of relational tables on the web have 20 or more rows (Cafarella et al., 2008), and these large tables present a challenge for current Transformer models, which are typically limited to 512 tokens. Here we propose MATE, a novel Transformer architecture designed to model the structure of web tables. MATE uses sparse attention in a way that allows heads to efficiently attend to either rows or columns in a table. This architecture scales linearly with respect to speed and memory, and can handle documents containing more than 8000 tokens with current accelerators. MATE also has a more appropriate inductive bias for tabular data, and sets a new state-of-the-art for three table reasoning datasets. For HybridQA (Chen et al., 2020b), a dataset that involves large documents containing tables, we improve the best prior result by 19 points.

Bhuwan Dhingra, Jeremy R. Cole, Julian Martin Eisenschlos et al.

Many facts come with an expiration date, from the name of the President to the basketball team Lebron James plays for. But language models (LMs) are trained on snapshots of data collected at a specific moment in time, and this can limit their utility, especially in the closed-book setting where the pretraining corpus must contain the facts the model should memorize. We introduce a diagnostic dataset aimed at probing LMs for factual knowledge that changes over time and highlight problems with LMs at either end of the spectrum -- those trained on specific slices of temporal data, as well as those trained on a wide range of temporal data. To mitigate these problems, we propose a simple technique for jointly modeling text with its timestamp. This improves memorization of seen facts from the training time period, as well as calibration on predictions about unseen facts from future time periods. We also show that models trained with temporal context can be efficiently "refreshed" as new data arrives, without the need for retraining from scratch.

Syrine Krichene, Thomas Müller, Julian Martin Eisenschlos

Transformer-based approaches have been successfully used to obtain state-of-the-art accuracy on natural language processing (NLP) tasks with semi-structured tables. These model architectures are typically deep, resulting in slow training and inference, especially for long inputs. To improve efficiency while maintaining a high accuracy, we propose a new architecture, DoT, a double transformer model, that decomposes the problem into two sub-tasks: A shallow pruning transformer that selects the top-K tokens, followed by a deep task-specific transformer that takes as input those K tokens. Additionally, we modify the task-specific attention to incorporate the pruning scores. The two transformers are jointly trained by optimizing the task-specific loss. We run experiments on three benchmarks, including entailment and question-answering. We show that for a small drop of accuracy, DoT improves training and inference time by at least 50%. We also show that the pruning transformer effectively selects relevant tokens enabling the end-to-end model to maintain similar accuracy as slower baseline models. Finally, we analyse the pruning and give some insight into its impact on the task model.

Thomas Müller, Julian Martin Eisenschlos, Syrine Krichene

We present the TAPAS contribution to the Shared Task on Statement Verification and Evidence Finding with Tables (SemEval 2021 Task 9, Wang et al. (2021)). SEM TAB FACT Task A is a classification task of recognizing if a statement is entailed, neutral or refuted by the content of a given table. We adopt the binary TAPAS model of Eisenschlos et al. (2020) to this task. We learn two binary classification models: A first model to predict if a statement is neutral or non-neutral and a second one to predict if it is entailed or refuted. As the shared task training set contains only entailed or refuted examples, we generate artificial neutral examples to train the first model. Both models are pre-trained using a MASKLM objective, intermediate counter-factual and synthetic data (Eisenschlos et al., 2020) and TABFACT (Chen et al., 2020), a large table entailment dataset. We find that the artificial neutral examples are somewhat effective at training the first model, achieving 68.03 test F1 versus the 60.47 of a majority baseline. For the second stage, we find that the pre-training on the intermediate data and TABFACT improves the results over MASKLM pre-training (68.03 vs 57.01).

Jonathan Herzig, Thomas Müller, Syrine Krichene et al.

Recent advances in open-domain QA have led to strong models based on dense retrieval, but only focused on retrieving textual passages. In this work, we tackle open-domain QA over tables for the first time, and show that retrieval can be improved by a retriever designed to handle tabular context. We present an effective pre-training procedure for our retriever and improve retrieval quality with mined hard negatives. As relevant datasets are missing, we extract a subset of Natural Questions (Kwiatkowski et al., 2019) into a Table QA dataset. We find that our retriever improves retrieval results from 72.0 to 81.1 recall@10 and end-to-end QA results from 33.8 to 37.7 exact match, over a BERT based retriever.

Julian Martin Eisenschlos, Syrine Krichene, Thomas Müller

Table entailment, the binary classification task of finding if a sentence is supported or refuted by the content of a table, requires parsing language and table structure as well as numerical and discrete reasoning. While there is extensive work on textual entailment, table entailment is less well studied. We adapt TAPAS (Herzig et al., 2020), a table-based BERT model, to recognize entailment. Motivated by the benefits of data augmentation, we create a balanced dataset of millions of automatically created training examples which are learned in an intermediate step prior to fine-tuning. This new data is not only useful for table entailment, but also for SQA (Iyyer et al., 2017), a sequential table QA task. To be able to use long examples as input of BERT models, we evaluate table pruning techniques as a pre-processing step to drastically improve the training and prediction efficiency at a moderate drop in accuracy. The different methods set the new state-of-the-art on the TabFact (Chen et al., 2020) and SQA datasets.

Jonathan Herzig, Paweł Krzysztof Nowak, Thomas Müller et al.

Answering natural language questions over tables is usually seen as a semantic parsing task. To alleviate the collection cost of full logical forms, one popular approach focuses on weak supervision consisting of denotations instead of logical forms. However, training semantic parsers from weak supervision poses difficulties, and in addition, the generated logical forms are only used as an intermediate step prior to retrieving the denotation. In this paper, we present TAPAS, an approach to question answering over tables without generating logical forms. TAPAS trains from weak supervision, and predicts the denotation by selecting table cells and optionally applying a corresponding aggregation operator to such selection. TAPAS extends BERT's architecture to encode tables as input, initializes from an effective joint pre-training of text segments and tables crawled from Wikipedia, and is trained end-to-end. We experiment with three different semantic parsing datasets, and find that TAPAS outperforms or rivals semantic parsing models by improving state-of-the-art accuracy on SQA from 55.1 to 67.2 and performing on par with the state-of-the-art on WIKISQL and WIKITQ, but with a simpler model architecture. We additionally find that transfer learning, which is trivial in our setting, from WIKISQL to WIKITQ, yields 48.7 accuracy, 4.2 points above the state-of-the-art.

Julian Martin Eisenschlos, Sebastian Ruder, Piotr Czapla et al.

Pretrained language models are promising particularly for low-resource languages as they only require unlabelled data. However, training existing models requires huge amounts of compute, while pretrained cross-lingual models often underperform on low-resource languages. We propose Multi-lingual language model Fine-Tuning (MultiFiT) to enable practitioners to train and fine-tune language models efficiently in their own language. In addition, we propose a zero-shot method using an existing pretrained cross-lingual model. We evaluate our methods on two widely used cross-lingual classification datasets where they outperform models pretrained on orders of magnitude more data and compute. We release all models and code.